Model Catalog
New & noteworthy speech and voice models curated for LA Studio. Ready to deploy locally or to cloud workflows.
LA Studio Picks
Qwen3-TTS Base 0.6B
by Qwen
Qwen3-TTS Base 1.7B
by Qwen
Qwen3-TTS CustomVoice 1.7B
by Qwen
All Models (15)
Qwen3-ASR 0.6B
Lightweight Qwen3-ASR model for offline multilingual speech transcription and audio Q&A, using the CrispASR runtime.
Qwen3-ASR 1.7B
Medium Qwen3-ASR model for offline multilingual speech transcription and audio Q&A, using the CrispASR runtime.
Qwen3-TTS Base 0.6B
Lightweight Qwen3-TTS base model for fast zero-shot voice cloning from reference audio. Requires 24kHz reference WAV and transcript.
Qwen3-TTS Base 1.7B
High-fidelity Qwen3-TTS base model for zero-shot voice cloning from reference audio. Provides better prosody and quality than 0.6B.
Qwen3-TTS CustomVoice 1.7B
Qwen3-TTS model specialized for preset speakers and style instructions. Optimized for 24kHz high-fidelity output.
Qwen3-TTS 1.7B VoiceDesign
Qwen3-TTS model specialized in zero-shot voice design from text instructions. Outputs high-fidelity 24kHz audio matching your style descriptions.
Kokoro Vietnamese
Fine-tuned Vietnamese Kokoro TTS packaged for LA Studio with a native Windows CPU runtime, ONNX Runtime inference, and bundled vig2p sidecar phonemization.
Kokoro
A compact open-weight TTS model with 82M parameters, installed in LA Studio through a Kokoro backbone GGUF plus a per-voice GGUF pack.
OmniVoice
A quality-first local speech model family for multilingual text to speech, prompt conditioning, and zero-shot voice cloning with reference audio.
VibeVoice
Microsoft VibeVoice Realtime 0.5B packaged as CrispASR GGUF for low-latency local text-to-speech with selectable preset voice prompts.
Nemotron-3.5 ASR Streaming 0.6B
NVIDIA's 0.6B FastConformer RNN-T ASR model, run locally through CrispASR v0.8.6 or later.
Whisper.cpp
Native C/C++ inference for OpenAI Whisper models converted to the custom ggml format. Supports multilingual transcription, automatic language detection and translation to English, with optional timestamp, TinyDiarize and Silero VAD workflows in upstream whisper.cpp.
VoxCPM2
Tokenizer-free 2B multilingual TTS model with 48kHz output, natural-language voice design, and zero-shot voice cloning via reference WAV.
VieNeu-TTS v2 Turbo
A lightweight, C++ bilingual Vietnamese-English TTS and voice cloning engine optimized for CPU execution.
VieNeu-TTS v3 Turbo
VieNeu-TTS v3 Turbo is a Vietnamese-English TTS model with preset voices, reference-audio cloning, and the v3_native C++ runtime pipeline.